US marketing agencies use generative AI
Forrester and the 4As reported broad use across creative content, research, media and SEO strategy, and internal work in June 2026.
Forrester + 4As / June 2026Client permission / Advertising agencies
Make AI-enabled agency work easier for clients to approve.
Petrichor turns one client account’s AI use into bounded workflows, current evidence, and a client-ready answer. Your teams can move faster without asking the brand for blind trust.
$150,000 flagship system · optional $25,000 account diagnostic
Do not send confidential client material by email.
The agency does not need another AI promise. The client needs a decision it can defend.
01The pressure
Adoption met scrutiny
AI is already in the work. Client permission is catching up.
Agency adoption and brand concern are now happening at the same time. That creates a practical question about what the client will permit, under which conditions, and on what evidence.
of surveyed brand owners had agency AI concerns
Legal, ethical, and reputational risk led the concerns. Contract terms, ownership, data, warranties, and indemnities were already moving into review.
WFA / September 2024These studies show market pressure. They do not establish that a client will approve a specific use or that an agency will buy Petrichor’s service.
02The gap
Policy is not account proof
A global policy cannot answer an account-specific question.
Principles matter. A brand still needs to know what operates in its work, who owns the decisions, what evidence supports the answer, and what happens when the facts change.
| Agency policy can state | Account assurance must show |
|---|---|
| Our principles | Included account, services, workflows, people, tools, data, partners, dates, and exclusions. |
| Our approved tools | Relevant terms, configurations, allowed data, output use, monitoring, and change state for this work. |
| Human review is required | Who reviews what, against which criteria, with which release authority and escalation path. |
| We follow disclosure rules | The materiality decision, client instruction, content credentials, handoff, and retained record. |
03The decision
Conditional permission
Give the client a decision it can actually make.
Petrichor supplies the boundary, operating system, and evidence. The client records its own permission, conditions, prohibited uses, responsibilities, expiry, and notification triggers.
04The workflows
Start with actual work
Bound the workflows closest to the client and the asset.
Select a workflow to see what a client-ready answer must cover. The first engagement starts where AI use, client scrutiny, and commercial consequence meet.
Strategy and insights
Show which sources, client plans, research inputs, and synthetic methods may enter approved tools. Record source validation, confidential-data rules, and the human judgment behind the recommendation.
05The trace
The evidence chain
Trace every client statement to operating reality.
If one link is missing, the answer is not ready for a client to rely on.
Statement
A precise answer about the AI use being requested. It names the work, conditions, exclusions, and human decisions instead of relying on an agency-wide promise.
06The output
The client approval pack
A concise answer the client can evaluate. A system that keeps it true.
The reusable core prevents reinvention. The client overlay preserves account-specific terms, choices, responsibilities, and limits.
Account and SOW boundary
Included services, workflows, teams, tools, data classes, partners, dates, and explicit exclusions.
Approved, conditional, and prohibited uses
A client-readable use position that teams can apply at the point of work.
Tool, model, and supplier record
Material terms, configurations, retention and training treatment, monitoring, and change state.
Client data and brand-asset controls
Rules for confidential strategy, first-party and audience data, brand systems, source material, and access.
Rights and provenance record
Sources, licenses, talent and likeness decisions, human contribution, content credentials, and retained history.
Human accountability model
Named creative, factual, legal, claims, brand, and release decisions with escalation paths.
Disclosure decision
The client instruction, materiality analysis, consumer-facing treatment, business handoff, and retained record.
Partner and platform conditions
Freelancers, production companies, data suppliers, media platforms, and other subprocessors.
Evidence and exception summary
Tests, approvals, reviewer checks, incidents, exceptions, dates, owners, and unresolved items.
Client permission record
The client decision, conditions, its own required controls, prohibited uses, expiry, and notification triggers.
Specimen limitation: This is an illustrative structure. It is not a certification, independent audit, legal opinion, client permission, or guarantee of acceptance.
07The presentation
Agency-owned / Source-visible
Make the agency look prepared without hiding how the work was built.
The client should see a coherent agency position and still know who designed the evidence method, what it covers, and what it does not prove.
01Default
Petrichor-branded
Petrichor prepares the readiness system and client pack. Our role, scope, date, evidence basis, and limitations stay visible.
02Client presentation
Co-branded
The agency presents the work with Petrichor. The format gives the agency ownership without disguising who designed the method and evidence structure.
03Attributed
Agency operating annex
The agency can use its own visual system for an operational annex when the Petrichor method, source, date, scope, and limits remain intact.
04Not included
Independent conclusion
Implementation is not certification. Petrichor does not let an agency present readiness work as an independent audit, legal opinion, or client approval.
08The value
A commercial operating case
Build an approval advantage without selling governance.
The case is based on the agency’s numbers and one real approval problem. It does not depend on invented win-rate lifts or fear-based breach averages.
01
Client work enabled
Covered account value, contribution margin, buyer-confirmed influence, and conservative attribution.
02
Assurance labor removed
AI riders, questionnaires, reviews, and approval events multiplied by current agency and client-review time.
03
Approved capacity
People, validated hours, working weeks, realizable value, and conservative adoption.
Account revenue is context, not proof of causation. Petrichor does not claim that the system retains a client or wins work without documented evidence and a defensible comparison.
09The engagement
One flagship / One optional front door
Build the full system when the account test earns it.
The flagship builds an assurance system the agency can operate, defend, and keep current. The optional Diagnostic establishes the first account boundary and economic case when those facts are not ready.
Flagship engagement
AI Assurance System
$150,000 fixed fee
- Delivery
- 20 to 22 weeks
- Coverage
- Up to 8 high-risk workflows
- Activation
- Up to 150 staff
Build the client pack, account overlays, evidence system, workflow controls, staff activation, executive challenge, response library, and review architecture.
Optional entry
Account Diagnostic
$25,000 fixed fee
- Delivery
- 4 weeks
- Boundary
- 1 client account
- Coverage
- Up to 3 workflows
Map the permission question, evidence gaps, operating boundary, draft client pack, and quantified case. Fully credited when the flagship is contracted within 30 days.
Investment gate
Proceed only when the buyer validates at least 4× conservative first-year value using the agency’s own operating data. A narrower price means a narrower boundary, not weaker controls.
10The fit
Qualify the problem
Built for agencies with real client scrutiny, not generic AI interest.
Strong fit
- Enterprise clients are issuing AI terms, questions, restrictions, or approval requests.
- AI already touches live client work, final assets, audience data, or media decisions.
- A material account, pitch, renewal, or delivery model is affected.
- An agency executive owns the work and a brand-side risk owner can test the evidence.
Not yet
- The need is limited to a policy template, staff course, or tool recommendation.
- There is no live or near-term enterprise-client permission problem.
- The agency wants a badge, legal opinion, safety claim, or approval guarantee.
- The conservative value case cannot clear the 4× investment gate.
11The references
Aligned / Not overclaimed
Use the industry’s rules and standards without confusing them for account proof.
The system can map relevant contract, disclosure, provenance, risk, and management-system requirements. The client-specific service boundary remains the object Petrichor is building.
IABAI Transparency and Disclosure Framework4AsMaster-services-agreement guidanceC2PAContent provenance specificationUSCOCopyright and artificial intelligence reportsNISTAI Risk Management FrameworkISO/IEC42001 AI management systems
This work does not: provide legal advice, determine copyrightability, clear rights, certify security or compliance, verify every output, eliminate AI risk, or guarantee that a client, regulator, insurer, platform, or procurement team will accept the agency’s position.
12Questions
A fast first read
Questions agency and client leaders ask first.
01What does the client receive?
The client receives a bounded approval pack: the account and workflow scope, allowed and prohibited uses, tool and data conditions, human responsibilities, current evidence, exceptions, expiry, and a record for its own permission decision.
02How is this different from an agency AI policy?
A policy states intent across the agency. AI Assurance connects one client’s conditions to the tools, data, partners, controls, evidence, and human decisions operating in the covered work.
03Does this certify the agency or guarantee client approval?
No. The engagement is readiness and implementation work. The client owns its decision, and Petrichor does not claim certification, legal compliance, safety, account retention, or guaranteed approval.
04Can the materials be co-branded?
Yes. The pack can be Petrichor-branded, co-branded, or formatted as an attributed agency operating annex. Petrichor’s role, source, date, scope, evidence basis, and limitations remain visible.
05How does Petrichor handle confidential client information?
The first conversation does not require a client name or confidential material. Any engagement defines access, retention, storage, review, and deletion requirements before sensitive evidence is collected.
06Does this replace agency or client counsel?
No. The work complements qualified advertising, privacy, intellectual-property, employment, cybersecurity, and regulatory advisers. Those advisers remain responsible for their conclusions.
07Which workflows should we start with?
Start where AI use, client scrutiny, and commercial consequence meet. For many agencies that is creative development or content production for one enterprise account, but the boundary is selected from the actual approval event.
08What happens when a tool, model, vendor term, or client instruction changes?
The system assigns expiry and material-change triggers. A relevant change reopens the affected evidence, control, and permission decision instead of leaving the old answer in circulation.
09What is the difference between the Diagnostic and the flagship?
The $150,000 flagship builds the operating system across up to eight high-risk workflows and 150 staff over 20 to 22 weeks. The optional $25,000 Diagnostic maps three workflows over four weeks when the agency must first establish the boundary, evidence gaps, and business case.
Start with a real account
Bring one client question and the workflows behind it.
We will test whether the account can be bounded, whether the evidence can support a client decision, and whether the business case earns the investment. Do not send confidential client material by email.